Real-Time Odds Latency Benchmark — Measuring The Numbers That Matter.
Every vendor claims “sub-second latency” and “real-time odds.” Few operators actually measure it. The ones who do discover that quoted numbers and observed numbers differ dramatically — vendor-reported “internal processing” latency is typically 2-3x faster than end-to-end latency operators actually experience. This guide covers the methodology operators should use to measure real latency, interpret percentile distributions, and compare vendors on reproducible data. Same approach we invite operators to run against our real-time odds feed.
source_ts → operator_rx
measured over 7-day window
end-to-end, not internal
Measure what you’ll actually experience.
Vendors quote latency numbers that look impressive. The problem is almost always methodology — vendor measurements capture “internal processing time” within their systems, excluding network transit, upstream data ingest latency, and operator-side delivery variance. Operators who buy on these numbers consistently discover real latency 2-3x higher in production, with corresponding margin impact that was invisible at evaluation time.
The right methodology measures end-to-end latency — from the moment an event happens in the real world (goal, point, VAR decision) to the moment the operator’s backend receives the updated odds. This includes vendor-internal processing, network transit, operator ingest, and queue latency. It’s the number that actually determines whether sharp accounts can arb your book during the brief windows between event and price update.
The four-step methodology below is how serious operators measure vendor latency honestly. Each step produces reproducible data; combined they generate the percentile distribution that distinguishes operator-grade feeds from marketing-grade ones. For the full benchmark approach including risk and payment dimensions, see our sportsbook API benchmark framework.
How to measure honestly.
Each step requires 1-3 hours of engineering time. Full methodology produces reproducible percentile data that holds up to vendor pushback and supports confident vendor selection.
Instrument Both Endpoints
Add timing instrumentation to source event (where real-world event is detected) and operator receiver (where odds update arrives in your backend). Use synchronized clocks (NTP) across both measurement points. Record every tick with microsecond precision timestamp.
Capture During Peak Load
Measure during actual peak traffic — EPL Saturday afternoon, NBA Sunday evening, Champions League nights. Vendor sandboxes often perform better than production because load is lower. 7-day capture window gets statistically meaningful data across varied conditions.
Compute Percentile Distribution
Report p50 (median — half of requests are faster), p95 (worst 5% start here — the sharp arb window), p99 (edge cases). Average/mean hides tail latency; percentiles expose it. Vendors who only quote averages are hiding bad p99 behaviour.
Compare Vendors On Same Methodology
Run identical measurement against your current vendor and shortlist candidates. Apples-to-apples comparison surfaces real deltas. Document methodology so vendors can’t dispute the numbers; results typically speak for themselves across the WSGaming and competitor comparisons.
Common latency questions.
What’s a good target p95 latency? +
Sub-400ms end-to-end for competitive operations. 400-700ms acceptable for lower-tier operations. Above 800ms costs real margin to sharp arb. WSGaming p95 target is 400ms; operators observe 380ms typical during peak events.
Why do p99 numbers matter? +
P99 is where the sharp accounts live. If 99% of your ticks are fast but 1% are 3 seconds late, that 1% is where syndicates concentrate attacks. Operators optimizing only for average/median latency get hollowed out at the tail.
How do I measure source event timing? +
For football: use broadcast timestamp of goal event. For NBA: official scorekeeper tick. For esports: game API event timestamp. Reference against vendor-provided event timestamp in odds payload. Difference = total latency including vendor processing.
What if vendor refuses benchmark cooperation? +
Red flag. Serious vendors welcome benchmarking because they’re confident in numbers. Vendors who deflect have gaps to hide. Document the refusal and factor into selection decision — probably not operator-grade.
How does WSGaming perform on this benchmark? +
Published p50/p95/p99 numbers available under NDA. 380ms p95 end-to-end during EPL Saturday peak is our standard benchmark. We welcome operator-run benchmarks and share methodology.
What about geographic latency variance? +
Platform multi-region deployment handles this. APAC operators measured from APAC PoP; EU operators from EU PoP. Transcontinental deployment adds 100-200ms; regional deployment stays sub-400ms. Measure from your actual operator region.
How often should I re-benchmark? +
Quarterly for established operators. Monthly during vendor evaluation. Immediately after any vendor claims of performance improvement. Measurement is cheap; dependency on outdated numbers is expensive.
How do I start benchmarking WSGaming? +
Sandbox same-day, benchmark pack within business day. Reference clients, measurement scripts, and methodology documentation included. Reach out to start.
Measure honest latency.
Benchmark pack with methodology, scripts, reference clients. Run against WSGaming and compare to your current vendor on identical methodology. Data wins over claims.